<h2>Must know</h2>
<ul>
<li><strong>OpenAI’s new AI-driven software development</strong>: OpenAI has introduced “Harness Engineering,” an internal AI-driven methodology that uses Codex agents to automate tasks like code generation and testing. A five-month experiment reportedly resulted in a million lines of code generated without manual writing, shifting engineers’ roles to designing environments and providing feedback. This approach aims to enforce architectural boundaries through mechanical rules and structural tests, with documentation serving as a machine-readable source of truth for the AI agents. <a href="https://app.daily.dev/posts/dPCNsAxds" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic and Figma streamline design-to-code</strong>: Anthropic and Figma have teamed up to integrate Claude Code, allowing developers to convert production code into editable Figma designs. This new workflow enables sending live pages to Figma files with a simple prompt, facilitating collaboration on design changes. Updated designs can then round-trip back to code via the Figma MCP server, aiming to streamline the often-clunky developer-designer handoff. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare R2 speeds up uploads with local writes</strong>: Cloudflare has launched “Local Uploads” for its R2 object storage, now in open beta. This feature aims to cut cross-region write latency by up to 75% by writing data to the nearest Cloudflare PoP, then asynchronously replicating it to the bucket’s home region. Objects become immediately accessible after the initial write, without waiting for full replication, and there are no additional costs beyond standard Class A operations. <a href="https://app.daily.dev/posts/XbGrpoJDs" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>TypeScript 6 paves the way for a Go rewrite</strong>: TypeScript 6 has been released as a transitional update, primarily to prepare for a planned rewrite of the compiler in Go, aiming to address performance bottlenecks. <a href="https://app.daily.dev/posts/tiMDdyFcG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS enables event-driven architectures for RDS SQL Server</strong>: AWS now allows Lambda function triggers directly from RDS for SQL Server database events, enabling event-driven architectures for SQL Server databases. <a href="https://app.daily.dev/posts/KzhIuYzeh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New LLM inference engine boasts significant speedups</strong>: NTransformer, a new C++/CUDA LLM inference engine, claims an 83x speedup for Llama 3.1 70B on a single RTX 3090, using adaptive caching and layer skipping. <a href="https://app.daily.dev/feed-by-ids?id=Ahx4zgQks&amp;id=DyVC0qyKG&amp;id=Fkac7nXWM&amp;id=K01E44V9h&amp;id=WoJMuyqAQ&amp;id=mulSXvfKt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 sees graphics and encryption updates</strong>: Linux 7.0 is getting initial DRM kernel graphics driver fixes for AMDGPU and renewed patch activity for eCryptfs, its per-directory encryption filesystem. <a href="https://app.daily.dev/feed-by-ids?id=2U7gOjLpG&amp;id=BZn5Av50T&amp;id=LutMi5Kcz&amp;id=MGUcuQgst&amp;id=gNaBG2osm&amp;id=u9gGdIsNO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GrapheneOS offers a privacy-centric Android experience</strong>: GrapheneOS, a privacy-focused Android alternative, offers a de-Googled experience with sandboxed Google Play as an option, though it comes with trade-offs like limited tap-to-pay. <a href="https://app.daily.dev/feed-by-ids?id=840Dw8MXb&amp;id=OwmBaq9Nm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloud skills still matter, but with an AI twist</strong>: A former AWS solutions architect suggests cloud skills remain valuable but require AI knowledge and a strong personal brand, noting all current AWS postings require on-site work. <a href="https://app.daily.dev/feed-by-ids?id=8yc6ZDNd6&amp;id=WAm3NKS9B&amp;id=djwfpUzVm&amp;id=gX2zbD1Ji" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>.NET 11 Preview 1 introduces async and compression upgrades</strong>: .NET 11 Preview 1 has been released, bringing runtime-level async improvements, Zstandard compression support, and new C# 15 language features. <a href="https://app.daily.dev/posts/QBAQQNpLN" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Altman says AI water usage claims are ‘untrue,’ compares AI to raising a child</strong>: OpenAI CEO Sam Altman, in a stunning display of PR wizardry, dismissed claims of high AI water usage as “completely untrue,” while simultaneously acknowledging total energy consumption is a legitimate concern. He then pivoted to comparing AI training costs to human development, because apparently, a ChatGPT query is now equivalent to raising a child for 20 years. TechCrunch dutifully reported this groundbreaking insight. <a href="https://app.daily.dev/feed-by-ids?id=95MvV75SN&amp;id=mGa3eARS3&amp;id=ogUHau6h0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gemini 3.1 Pro: benchmark king, real-world flop?</strong>: Google’s Gemini 3.1 Pro is apparently a benchmark champion, scoring 78% on ARC AGI 2 and 100% on a skateboarding spatial reasoning test. However, according to some, this “benchmaxing” comes at the cost of real-world usability, with reports of unreliable tool calling, looping behavior, and hallucinated packages. Because who needs a functional AI when you have impressive, albeit irrelevant, benchmark scores? <a href="https://app.daily.dev/feed-by-ids?id=95MvV75SN&amp;id=mGa3eARS3&amp;id=ogUHau6h0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents: the next big security nightmare</strong>: A BBC reporter was allegedly hacked via Orchards, an “AI vibe coding platform,” after a security researcher injected malicious code. Meanwhile, OpenClaw, an open-source AI agent, revealed 512 vulnerabilities, including a critical remote code execution flaw, with 12-20% of its marketplace packages being malicious. It seems the future of AI agents is less about superintelligence and more about super-security risks, as detailed in various articles. <a href="https://app.daily.dev/feed-by-ids?id=1HpSP7L9S&amp;id=1TelNKsBD&amp;id=9xLrRFK2b&amp;id=FuKqPJxtz&amp;id=Jm3weKMWA&amp;id=ZmcjIu8ai&amp;id=em7afQ5sm&amp;id=kdJWfknAe&amp;id=ohhVPexbM" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
